Corruption in aid programs is a cyclical topic. Every scandal generates headlines, political reaction, tighter controls and then, usually, silence until the next scandal erupts. Such cycles are not helpful and we never really find out if such corruption is large and systematic or small and isolated.
Last year I commented on press exaggerations of a Global Fund investigation into the use of grants in Mauritania. The AP article made irresponsible allegations about the overall portfolio of the Global Fund on the basis of reports from a few small projects. (The Global Fund deserves credit for publishing its investigative reports with transparency). I argued that the only way out of this dysfunctional cycle was to have a way to judge how representative such investigations are of the Global Fund’s overall portfolio of projects.
Last week, we convened a workshop at CGD to ask whether it would be feasible for aid agencies to generate such “portfolio estimates” of corruption (or improper payments more broadly). This would require allocating investigative resources to a representative sample of projects on an annual basis instead of focusing investigations exclusively on cases where abuses were alleged. With a small group of high-level aid agency officials, inspectors general, researchers, and former police officers, we discussed examples of institutions that regularly produce such portfolio estimates and considered the unique features of foreign aid that might facilitate or obstruct such an approach.
The clearest story came from the New York Police Department (NYPD) which, in the early 1990s, started a program of targeted integrity testing. When officers were accused of malfeasance, investigators would test their honesty by presenting them with an artificial (and recorded) opportunity to take a bribe or to profit from their position. This process led to dismissals and/or convictions of about 25% of the officers who were tested in this way.
However targeted testing only revealed how many officers were likely to be corrupt once someone had lodged a complaint. Such targeted testing is important, but it cannot answer the question the public really wants to know which is: how many officers have the integrity to turn down (and report) an attempted bribe? Today the NYPD has the answer to this question because since 1996 they have also been conducting random testing. Officers know that they may be randomly selected and presented with an opportunity to commit a corrupt act in the course of their work. About 3 to 4 percent of those who are randomly tested get caught in some infraction and are dismissed and/or convicted through this process. In other words, as a result of this random testing, the NYPD Commissioner can confidently tell New York City residents that 96 to 97 percent of the 34,500 police officers at work are honest.
The NYPD’s random testing is the clearest example of a portfolio estimate, but we also discussed other cases. The United States Improper Payments Elimination and Recovery Act (IPERA) requires agencies to provide portfolio estimates for high-risk programs. Thus, US agencies provide annual figures on improper payments (which include simple errors as well as fraud) under Medicare Fee-for-Service ($28.8 billion or 8.6% in 2011), the Earned Income Tax Credit ($15.2 Billion or 23.5% in 2011), Defense programs ($900 million in 2011), and many other programs. The methods for generating such estimates vary considerably – some are quite good while others are questionable. Similar variability in the quality of estimates has been found in international comparisons of fraud in social security systems. We also discussed a report that alleged corruption in the US aid portfolio to Afghanistan, with participants pointing out that the methodology for obtaining those figures was never explained.
Having a portfolio estimate for improper payments or fraud is important for judging whether the results of any particular investigation are an aberration or a systematic problem. But there are other benefits:
- A deterrent effect – when people know that there is a chance of being randomly selected for an investigation, it encourages them to behave honestly.
- A management effect – a common response to scandals is to introduce more stringent procurement rules and reporting requirements without evidence that this would make a difference, yet regular portfolio estimates would help managers to judge if their strategies are succeeding.
- A resource allocation effect – reliable information about the types and location of fraud and errors can help investigative offices prioritize their limited resources and help agency heads decide how much to dedicate to investigation.
- A discovery effect – current systems tend to discover fraud and errors only when a whistleblower or disgruntled competitor raises the alarm; portfolio estimates from random testing can uncover scams that would not otherwise be revealed.
The benefits are huge but two big questions remain: Are portfolio estimates feasible? And if they are feasible, what would it take for aid agencies to try? Our next steps are to answer these questions and maintain an open dialogue with the agencies involved.